Speaker
Description
Weakly-bound complexes are very appealing for experimental investigations of resonances in dissociation dynamics, which is of vital importance to roaming reactions. Planning and elucidating experiments requires accurate quantum mechanical calculations of (ro-)vibrational energies up to dissociation, which is a challenging task for these systems because of their flexible degrees of freedom and large configuration space. Standard predictions for these problems represent the wavefunctions as a linear combination of some fixed basis set. The quality of the predictions highly depend on the choice of the basis set and the computational costs scale poorly with the dimension of the problem and the number of excited states considered.
To address these problems, we present a nonlinear variational principle that approximates molecular states in the linear span of augmented basis sets. These sets are constructed using normalising flows where the base distributions are the functions of a standard basis set of $L^2$. The proposed framework shows more stability during training than nonlinear calculations using standard neural networks. It promises to mitigate the curse of dimensionality and to allow for more accurate computations of excited states. We present simulations on the water molecule and convergence guarantees under certain assumptions on the potential-energy surface. Moreover, a perspective to use these methods for (ro-)vibrational and dynamics calculations of weakly-bound complexes is presented.